Citation: BoneKEy Reports 2, Article number: 386 (2013) | doi:10.1038/bonekey.2013.

120
& 2013 International Bone & Mineral Society All rights reserved 2047-6396/13
www.nature.com/bonekey

REVIEW

Finite element analysis for prediction of bone

strength
Philippe K Zysset1, Enrico DallAra2, Peter Varga3 and Dieter H Pahr4
1
Institute for Surgical Technology and Biomechanics, University of Bern, Bern, Switzerland. 2Department of Mechanical
Engineering, University of Sheffield, Sheffield, UK. 3Julius Wolff Institut, Charite - Universitatsmedizin Berlin, Berlin,
Germany. 4Institute of Lightweight Design and Structural Biomechanics, Vienna University of Technology, Wien, Austria.

Finite element (FE) analysis has been applied for the past 40 years to simulate the mechanical behavior of bone. Although
several validation studies have been performed on specific anatomical sites and load cases, this study aims to review the
predictability of human bone strength at the three major osteoporotic fracture sites quantified in recently completed
in vitro studies at our former institute. Specifically, the performance of FE analysis based on clinical computer
tomography (QCT) is compared with the ones of the current densitometric standards, bone mineral content, bone
mineral density (BMD) and areal BMD (aBMD). Clinical fractures were produced in monotonic axial compression of the
distal radii, vertebral sections and in side loading of the proximal femora. QCT-based FE models of the three bones were
developed to simulate as closely as possible the boundary conditions of each experiment. For all sites, the FE
methodology exhibited the lowest errors and the highest correlations in predicting the experimental bone strength.
Likely due to the improved CT image resolution, the quality of the FE prediction in the peripheral skeleton using
high-resolution peripheral CT was superior to that in the axial skeleton with whole-body QCT. Because of its projective
and scalar nature, the performance of aBMD in predicting bone strength depended on loading mode and was
significantly inferior to FE in axial compression of radial or vertebral sections but not significantly inferior to FE in side
loading of the femur. Considering the cumulated evidence from the published validation studies, it is concluded that FE
models provide the most reliable surrogates of bone strength at any of the three fracture sites.

BoneKEy Reports 2, Article number: 386 (2013) | doi:10.1038/bonekey.2013.120

Introduction
Application of finite element (FE) analysis to determine stresses
in human bones was initiated in the early 1970s in the field of
orthopedic biomechanics.
In the 1980s, FE analysis was already perceived as a highly
promising tool in orthopedics and other clinical fields, but the
enormous gap between the idealized biomechanical models
that could be solved by FE and the clinical applications was
readily recognized. The question as to whether this new
computational methodology helps the patient could not be
answered definitively. The communication between engi-
neering and the medical faculty was considered a major
problem that contributed to the establishment of joint edu-
cational programs.
The lack of accurate three-dimensional (3D) anatomy, bone
material properties and loading conditions was also identified.
In fact, the FE method was available but the necessary input
data were missing.
In the 1990s, the availability of computer tomography (CT)
provided a strong impulse for the generation of anatomy-
specific FE models of whole bones.1 Bone heterogeneity and
anisotropy were found to be important for both static and
dynamic analyses.2
The elastic properties of cortical bone were assembled from
those of osteons by homogenization theory.3 Trabecular
bone was analyzed by FE as idealized open and/or closed cell
models or from the first microCT reconstructions,4 and its
apparent elastic properties were also computed by homo-
genization theory.5 Nonlinear 3D constitutive models for bone
based on plasticity and damage appeared in the literature.6
Bone-remodeling scenarios were also intensively developed
using FE.7,8

Correspondence: Professor PK Zysset, Institute for Surgical Technology and Biomechanics, University of Bern, Stauffacherstrasse 78, Bern CH-3014, Switzerland.
E-mail: philippe.zysset@istb.unibe.ch
Received 3 January 2013; accepted 25 June 2013; published online 7 August 2013

BoneKEy Reports | AUGUST 2013 1

 FE analysis to predict bone strength
PK Zysset et al
During the past Bone and Joint decade (20002010),
awareness of osteoporosis increased and the number of
publications with the string FE grew rapidly.9 The development
of computer hardware and software tools allowed substantial
progress in the area of model validation10 and began the era of
patient-specific FE modeling.
A recent report reviewed the principles of the CT-based FE
method and its application in the hip and spine.11 In this paper,
we aim to synthesize three recent validation experiments
performed in our laboratory for patient-specific modeling of
three skeletal sites at high risk of osteoporotic fractures: the
distal radius, the vertebral body and the proximal femur. In
particular, we quantify the predictive power of the voxel-based
FE methodology for the image resolution available in vivo and
compare it with the current densitometric surrogates of bone
strength, bone mineral content (BMC) and areal bone mineral
density (aBMD), for each anatomical site.
Study Design
The adopted study design consisted first in obtaining a sub-
stantial number of human bones from both genders and a broad
age range with the approval of the local ethics commission.
Custom chambers were built to scan the bones in a predefined
coordinate system with CT. Before scanning, the bones were
immersed in saline solution to mimic the presence of soft issues
and exposed to vacuum in order to remove air bubbles that
create artifacts in the local mineral density. A calibration
phantom was used to calculate the calibration function between
the attenuation coefficient and BMD for each bone.
Representative but maximally simplified loading scenarios
were then applied to each bone to reproduce typical fractures
observed in patients with osteoporosis. The number of
materials and interfaces intervening in the testing set-up were
minimized for proper reproduction of the loading scenario in the
FE model. The bones were tested monotonically to failure in a
servohydraulic testing system (MTS 858 Mini-Bionix, MTS
Corp, Eden Prairie, MN, USA).
Voxel meshes of the bones and potential embedding
materials were generated in a fully automated manner with an
element size that saved computational resources in the per-
spective of large clinical studies but showed acceptable
accuracy when compared with refined meshes.12 For micro-
structural FE (microFE) models such as those obtained from
high-resolution peripheral CT (HR-pQCT), the elastic properties
of bone tissue were based on nanoindentation measure-
ments.13 For homogenized FE models generated from clinical
CT (QCT), the nonlinear material properties of each element
were based on the local calibrated BMD converted into bone
volume fraction and morphologymechanical property rela-
tionships for trabecular bone14 that were identified on human
biopsies.15 When available, fabric information was included in
the morphologymechanical property relationships, but the
cortex was not distinguished from trabecular bone.
The morphologymechanical property relationships fitted for
trabecular bone were adequately extrapolated to reach bone
tissue properties for a volume fraction equal to 1. The con-
stitutive model for bone tissue was elastic, plastic with damage,
that is, coupled accumulation of plastic deformation and
reduction of stiffness beyond a given yield surface in stress
space.16 The FE meshes were generated with a custom script
2 AUGUST 2013 | www.nature.com/bonekey
manager (MedTool, TU-Wien, Vienna, Austria) and the analyses
were performed on a computing server (2  2 XEON 3.0 GHz
processors, 32 GB RAM) with ParFE (Professor PArbenz, ETHZ,
Zurich, Switzerland) software for the linear models of the distal
radius and with Abaqus (version 6.8, Simulia, Dassault Sys-
temes, Velizy-Villacoublay, France) software for the nonlinear
models of the vertebral sections and proximal femora.
Full methodological details of the three studies on the distal
radius, vertebral sections and proximal femora can be found in
Varga et al.17 and DallAra et al.18,19
Linear regressions were calculated between the experi-
mental maximal force or strength and the FE or densitometric
variables. The adjusted Pearsons correlation coefficients and
the root mean square errors (RMSE) of the predictions were
calculated for all linear regressions. The statistical comparisons
of the dependent correlation coefficients were made with the T2
distribution described in Steiger.20
Ultimate Load of the Distal Radius Using HR-pQCT
Colles fracture is considered as an early osteoporotic fracture
and is characterized by an angulated compressive failure of the
distal radius.21 In a recent radiographical study, the fracture line
was located dorsally 7.92.7 mm and palmarly 11.73.9 mm
proximal to the corresponding lunate fossa apex.22
Materials and methods. After alignment of the shaft and
embedding of the proximal and distal ends in polymethyl
methacrylate (PMMA), 26 formalin-fixed distal radii (11 men
and 16 women, aged 61103) were scanned using a clinical HR-
pQCT (XtremeCT, Scanco Medical, Bruttisellen, Switzerland)
protocol with a voxel size of 82 m. The biomechanical testing
set-up developed in Varga et al.23 and illustrated in Figure 1
consisted of an axial loading of the radial shaft with an inclined
distal embedding that produced a combined compression and
bending in the distal region. All bones were tested to failure with a
loading rate of 1 mm s  1. Cutting of the radii in a sagittal plane
after testing revealed a white, inclined compression line that
corresponded rather well to the definition of Colles fracture.
Although a nonlinear homogenized FE model of this experiment
was realized and allowed to reproduce the individual fracture
loads and locations of this experiment, a clinically more
important question was addressed in a subsequent paper24
where only the distal 9 mm section of the radius corresponding to
the standard clinical protocol of the HR-pQCT manufacturer was
modeled (Figure 1). As Colles fracture was found to occur in the
vicinity of this section, it was legitimate to test whether the
strength of that standard section would predict the strength of the
whole-distal radius. A simple linear microFE model was therefore
generated for each radial section and subjected to axial
compression between two rigid plates with infinite friction. The
elastic modulus of bone tissue was set to 15 GPa and the Poisson
ratio to 0.3, based on a previous indentation study.23,25 The
fracture load was determined from the linear FE analysis using an
empirical rule proposed by Pistoia et al.26
Results. The regression plots for prediction of the experi-
mental strength of the distal radius are shown in Figure 2
for microFE, BMC and simulated aBMD. The predictions of
microFE are quantitatively correct with a slope close to 1.
Although the microFE analysis delivers the highest correlation

Figure 1 On the left, experimental model of Colles fracture.23 On the right, FE model of the standard 9 mm radius section obtained with the clinical protocol of the HR-pQCT.24
The boundary conditions correspond to a displacement-based axial compression at the distal end with fully fixed nodes at the proximal end of the section.
coefficient and the lowest RMSE for maximal force, the pre-
dictions of BMC and CSA*BMD (cross-sectional area times
bone mineral density) are remarkable and not significantly
different (Tables 1 and 2). The predictions of aBMD are sig-
nificantly inferior for maximal load (P 0.003) but not for
maximal apparent stress (P 0.284). The predictions of BMD
are approaching those of the microFE method for maximal
apparent stress (P 0.347).
Discussion. As shown in Table 3, the precise biomechanical
testing set-up of the present study allowed achieving better
microFE predictions of maximal force of the standard distal
radius section than in the studies by Pistoia et al.26 (R2 0.75)
and Mueller et al.27 (R2 0.73). The experimental compression
of the wrist including soft tissues but neglecting muscle forces
may have generated stress fields that were more difficult to
reproduce in a FE model. Given the HR-pQCT images, the use
of a homogenized nonlinear model in Varga et al.28 did not
improve the correlations with respect to linear microFE models
for radial sections but may help to reduce the necessary
computational resources for clinical applications in the future.
The remarkable predictions of maximum force by BMC may be
attributed to the optimal structure of the distal radius to sustain
axial compressive load. There is no superfluous tissue and
mineral mass correlates significantly with stiffness, which in turn
is closely related to strength. The weak performance of aBMD in
predicting maximal force was also observed by Pistoia et al.26
The implications of these outstanding in vitro results of
microFE analysis are reflected by its discrimination power of
fragility fractures in a number of cross-sectional studies, for
example, Vilayphiou et al.,29 Stein et al.,30 Vilayphiou et al.,31
Boutroy et al.32 and Melton et al.33
Ultimate Load of Vertebral Sections Using QCT
Nontraumatic vertebral fractures represent a hallmark of
osteoporosis and have been thoroughly classified.34 Although
numerous biomechanical tests were realized by embedding the
vertebral bodies in PMMA (Table 3), an alternative consisted in
BoneKEy Reports | AUGUST 2013
FE analysis to predict bone strength
PK Zysset et al
resecting the endplates to produce two flat surfaces that adapt
well to compression plates. A recent FE analysis study
demonstrated that the strength of vertebral bodies embedded
in PMMA is essentially equivalent to that of the corresponding
vertebral section without the endplates.35
Materials and methods. In the second study performed in our
laboratory,18 37 vertebral sections (7 men and 3 women, aged
4482, T12-L5) were prepared and scanned in a custom
chamber using a clinical CT protocol (Brilliance64, Philips,
Hamburg, Germany) with a voxel size of 0.391  0.391 
0.45 mm3. The sections were compressed monotonically at a
rate of 5 mm min  1 up to failure between two steel plates with a
ball joint above the upper plate to allow free rotation (Figure 3).
The vertebral bodies were positioned carefully to bring the axis
of the testing machine anterior to the center of gravity of the
middle cross-section. Most often, anterior, and twice posterior,
wedge fractures were produced, which can be found in
radiological assessment schemes.
A homogenized voxel-based FE model of each section was
generated with an edge size of 1.3 mm. The nonlinear material
properties were transverse isotropic with the axis of symmetry
along the inferiorsuperior direction. The motion of the upper
plate measured in the experiment was imposed on the top
surface of the FE model. Like in the experiment, the fracture
load of the FE analysis was obtained from the maximum of the
forcedisplacement curve.
Results. The regression graphs for prediction of the in vitro
strength of the vertebral sections are shown in Figure 4 for
homogenized FE (voxelFE), BMC and lateral aBMD. The linear
regression of the FE analysis is close to the 1:1 relationship
without fitting of the bone material properties. The FE analysis
provides the highest correlations and the minimum relative
errors. The difference between the FE model and BMC or
CSABMD in predicting maximal force is more pronounced than
that for the radial sections but remains insignificant (P 0.175
and P 0.179, respectively). The lowest correlations were
obtained by aBMD and the differences with voxelFE did reach
3
 FE analysis to predict bone strength
PK Zysset et al

significance for both maximum force and maximum apparent
stress (Table 1). The predictions of BMD are again comparable
to those of voxelFE for maximum apparent strength (P 0.283).
Discussion. As shown in Table 3, other validation studies on
vertebral bodies achieved comparable or better correlations
8 pQCT.41 A further issue is the absence of any fitting in the
adj. R2=0.92
Exp Maximum Force [kN]
with maximum force using QCT-based homogenized FE, for
example, Chevalier et al.,36 Imai et al.,37 Buckley et al.,38
Crawford et al.39 and Martin et al.40 Size and heterogeneity of
the data set are known to affect such correlations. Never-
theless, modeling of the cortex in Imai et al.37 seems to
substantially improve the FE predictions in the vertebral body.
This effect was also reported in an ex vivo study using HR-
FE model by DallAra et al.18 Tuning material properties for a

6 given data set may well achieve higher correlations but dis-
regard the universality sought for a better surrogate of bone
strength. The positive results of these biomechanical studies
4 in vitro provided solid ground for its use in cross-sectional
or prospective clinical studies, for example, Imai et al.,42
2 Graeff et al.43 and Keaveny et al.44 Finally, nonlinear FE of the
RMSE=13.6%
vertebral body was shown to be statistically more sensitive
0
than densitometric variables to evaluate the biomechanical
0 2 4 6 8 benefit of drugs against osteoporosis.45
vFE Maximum Force [kN]

8 Ultimate load of the Proximal Femur in a Sideways Fall

adj. R2=0.892 Configuration Using QCT
Exp Maximum Force [kN]

6 Fractures of the hip lead to the highest mortality, morbidity and

health-care costs. A very large body of research has been
4 conducted to study the mechanical behavior of the proximal
femur and specifically for the validation of FE models in terms
of strain,10,46,47 fracture load4851 and fracture location.52,53
2 Although very few fractures seem to occur in the standing
RMSE=15.8% position, both stance and side-loading configurations were
0 investigated.
0.0 0.2 0.4 0.6 0.8 1.0 1.2
BMC [g]
Materials and methods. In the third study conducted in our
8 laboratory,19 36 pairs of proximal femora (17 men and 19 women,
adj. R2=0.769 aged 4696) were freed of soft tissues except the cartilage of the
Exp Maximum Force [kN]

head and embedded distally in polyurethane (PU). Each bone

6
was submerged in a custom chamber filled with saline and
subjected to a standard dual-energy X-ray absorptiometry
4 examination (Discovery QDR, Hologic Inc., Bedford, MA, USA).
In the same chamber, each sample was then scanned by clinical
2 QCT (Brilliance64, Philips) with a calibration phantom. The
reconstruction voxel size was 0.33  0.33  1.0 mm3.
RMSE=23.1%
0
The left and right femora of each pair were randomly assigned
0.0 0.1 0.2 0.3 0.4 to a stance and side group for assignment to the biomechanical
aBMD [g/cm2] loading protocol. The side configuration did mimic a sideways
fall onto the greater trochanter. The distal embedding of each
Figure 2 Linear regressions between experimental strength of the distal radii and
microFE analysis (MicroFE), bone mineral content (BMC) and areal bone mineral femur was fully fixed in a servohydraulic testing system, and an
density (aBMD). The blue solid line is the regression, the red dotted lines are the axial load was applied to the head through an embedding cap in
confidence intervals and the black dashed line is the 1:1 relationship. Adj. R 2, the PU that allowed rotations. Custom bearings eliminated trans-
adjusted coefficient of determination; RMSE, relative root mean square error. verse forces and torsional moment. Each bone was loaded

Table 1 Adjusted correlations coefficients for prediction of maximal force and maximal apparent stress using FE, QCT and aBMD

Adj. R2 Maximum force Maximum apparent stress

Site Imaging N FE BMD*CSA BMC aBMD BMD FE

Distal radius HR-pQCT 21 0.920 0.893 (P 0.158) 0.892 (P 0.151) 0.769* (P 0.003) 0.707 (P 0.284) 0.751 (P 0.347) 0.775
Vertebra QCT 37 0.768 0.707 (P 0.179) 0.692 (P 0.175) 0.624* (P 0.046) 0.467* (P 0.001) 0.732 (P 0.283) 0.771
Hip (fall) QCT 36 0.824 0.699* (P 0.032) 0.791 (P 0.264)
Abbreviations: aBMD, areal bone mineral density; Adj. R2, adjusted coefficient of determination; BMC, bone mineral content; BMD, bone mineral density; FE, finite
element; HR-pQCT, high-resolution peripheral computer tomography. For aBMD, which is neither a structural nor a material variable, the predictions of both maximal
force and apparent stress are reported. *Designates a statistically different correlation coefficient with respect to the corresponding FE variable (Po0.05).

4 AUGUST 2013 | www.nature.com/bonekey

 FE analysis to predict bone strength
PK Zysset et al

Figure 3 On the left, mechanical testing set-up of a human vertebral section. The upper plate is free to rotate and its position during the test is recorded with the help of three linear
variable differential transformers (LVDT). On the right, a voxel-based homogenized FE model of a vertebral section is shown. The boundary conditions correspond to a displacement-
driven axial compression at the cranial end with free rotation and fully fixed nodes at the caudal end. The gray level corresponds to the calibrated bone volume fraction.

quasistatically and monotonically to failure at a displacement 10

rate of 5mm min  1. The fracture pattern of each bone was then adj. R2=0.768

Exp Maximum Force [kN]

assessed with an antero-posterior (AP) X-ray. A homogenized 8
voxel-based FE model of each bone with the relevant embed-
ding was created with an element size of 3  3  3 mm3 (Figure 6
5). The nonlinear material properties were isotropic. The distal
end was fully fixed and the embedding of the greater trochanter 4
was free to move in the horizontal plane. The axial motion of the
2
experimental loading device was applied to the upper surface of RMSE=14.8%
the embedding cap taken as a rigid body where the other
0
degrees of freedom were kept free. The fracture load of both 0 2 4 6 8 10
experiment and FE analysis was again defined as the maximum VoxelFE Maximum Force [kN]
of the forcedisplacement curve.
10
Exp Maximum Force [kN]

adj. R2=0.692
Results. In general, the experiments produced typical sub- 8
capital, neck and trochanteric fractures with a few atypical or
indiscernible fractures. The agreement between X-rays and 6
damage distribution computed by FE was 71% for the side
configuration. The regression graphs of prediction of the 4
femoral fracture load are shown in Figure 6 for homogenized FE
(VoxelFE), BMC and the aBMD of the neck. There is a fair 2
RMSE=17.0%
quantitative correspondence between the experiment and the
0
FE prediction for the fall configuration. The coefficient of 0 2 4 6 8
determination of the FE prediction of maximum force is higher BMC [g]
than that of aBMD, but the difference did not show statistical
significance (P 0.264). In contrast to the radial and vertebral 10
adj. R2=0.624
Exp Maximum Force [kN]

sections, the predictions of maximum force by BMC in the femur

8
were significantly lower than those of FE (P 0.032).
6
Discussion. The correlation coefficients between the FE
strength predictions and the experimental strength in the fall 4
configuration are highly consistent among the validation studies
listed in Table 3. Surprisingly, the inclusion of the cortex in the 2
proximal femur does not appear to have the same impact on the RMSE=18.8%
0
strength predictions as in the vertebral body. This may be related 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
to the difficulty in extracting the varying thickness of the cortex aBMD [g/cm2]
and material anisotropy of the proximal femur from whole-body
QCT images. It should be noted that for homogenized voxel Figure 4 Linear regressions between experimental strength of the vertebral
sections and homogenized FE analysis (voxelFE), bone mineral content (BMC)
models of the proximal femur, nonlinear analysis performs better
and areal bone mineral density (aBMD). The blue solid line is the regression, the red
than linear analysis.49 The predictions of maximum force by the dotted lines are the confidence intervals and the black dashed line is the 1:1
FE method were higher but not significantly different than that by relationship. Adj. R 2, the adjusted coefficient of determination; RMSE, relative root
aBMD, which may reflect a lack of statistical power but also mean square error.

BoneKEy Reports | AUGUST 2013 5

 FE analysis to predict bone strength
PK Zysset et al

Figure 5 On the left, experimental design for testing the human proximal femur in a side-loading configuration. On the right, voxel-based FE models of the human proximal femur
in the same side-loading configuration. The gray level represents the bone volume fraction.

confirm the value of neck aBMD in predicting not only hip

5 adj. R2=0.824

Exp Maximum Force [kN]

strength in a side configuration but also hip fractures, as shown
in numerous epidemiological studies, for example, Johnell 4
et al.54 Nevertheless, one study could demonstrate higher
correlation coefficients for some FE strength predictions when 3
compared with hip aBMD in the fall configuration.55
The experimental evidence supporting these QCT-based FE 2
models of the proximal femur motivated their application in
several clinical and space-flight studies.5662 Finally, a recent 1
RMSE=14.6%
study in White women and men showed that femoral strength
0
assessed by homogenized FE declines much faster with age 0 1 2 3 4 5
than aBMD.63 VoxelFE Maximum Force [kN]

5 adj. R2=0.699
Exp Maximum Force [kN]

General discussion
Three biomechanical experiments were developed to repro- 4
duce the three most common osteoporotic fractures in the
3
laboratory and to compare the performance of several
surrogates of the bone strength. 2
A recurrent criticism of such experiments is that they are only
caricatures of the load cases occurring in the real lives of real 1
patients. However, the aim of these experiments is not the RMSE=19.1%
perfect match with an individual reality but the systematic 0
0 1 2 3 4 5 6
validation of the QCT-based FE methodology for representative
BMC [g/cm3]]
anatomies and loading modes. The FE method is founded on
the universal principles of mechanics and has the potential,
5 adj. R2=0.791
Exp Maximum Force [kN]

after proper validation, to simulate arbitrarily complex loading

cases whenever these are accurately known, for instance, from 4
musculoskeletal models of the human body.64 The particular
load cases selected in these experiments should, therefore, not 3
be seen as limitations in regard to the diversity of clinical
fractures, but rather as opportunities to test the universal 2
assumptions of the FE approach.
The highest correlations and lowest errors for prediction of 1
RMSE=15.9%
the maximum force using the FE method were obtained at the
0
distal radius site that benefits from detailed HR-pQCT images 0.0 0.2 0.4 0.6 0.8 1.0
and for which microFE models can be realized. The FE aBMD [g/cm2]
models for vertebral sections and proximal femora exhibited
Figure 6 Linear regressions between experimental strength of the proximal femora
weaker predictions, as their bone material properties were and homogenized FE analysis (voxelFE), bone mineral content (BMC) and areal bone
homogenized and based on QCT images with comparably mineral density (aBMD). The blue solid line is the regression, the red dotted lines are the
coarser resolution. The proper consideration of cortical confidence intervals and the black dashed line is the 1:1 relationship. Adj. R 2, adjusted
thickness and trabecular anisotropy may be the key factors coefficient of determination; RMSE, relative root mean square error.

6 AUGUST 2013 | www.nature.com/bonekey

 FE analysis to predict bone strength
PK Zysset et al

to improve the current models and are investigated in intensive (as opposed to extensive or additive) variables that
several laboratories using high-resolution CT.41,65,66 Given the normalize the bone mass by volume (BMD), the forces by the
excellent FE predictions obtained in the trabecular bone area (apparent stresses) and the displacements by height
samples using microCT,67 future improvements of CT tech- (apparent strains). In this perspective, aBMD is a hybrid variable
nology are expected to lead to improved FE models alongside that lies somewhere between BMC and BMD. In the case of the
the development of refined image-processing tools. side loading of the proximal femur, the resulting stresses are
In the case of axial compression, which was used for the distal heterogeneous and do not allow for a simple geometrical
radius and the vertebral sections, it is possible to define normalization.
In this context, the correlation coefficients (R2) and the
relative RMSE for the prediction of maximal force and maximal
apparent stress are summarized in Tables 1 and 2. Clearly, it
Table 2 Root mean square error of the prediction in percentage of the average makes sense to correlate intensive or extensive variables
experimental force or stress variable using FE, QCT and DXA
among themselves but not to mix them up. In the distal radius,
Adj. R2 Maximum force Maximum the extensive variables provided better correlations and lower
apparent stress errors than the intensive variables. In the vertebral sections,
they were comparable. The origin of this difference may be
Site Imaging N FE BMD*CSA BMC aBMD BMD FE because of the distinct coefficients of variation in maximal force
(32 versus 40%) and ultimate stress (31 versus 30%) in the
Distal HR- 21 13.6 15.8 15.8 23.1 18.2 16.8 16.0 vertebral body versus the distal radius. The normalized vari-
radius pQCT ables are attractive for moving from bone strength to a factor of
Vertebra QCT 37 14.8 16.6 17.0 18.8 22.0 15.7 14.5 risk toward bone fracture risk,68 which depends on the applied
Hip (fall) QCT 36 14.6 19.1 15.9
load and therefore on patient body weight. On the basis of a
Abbreviations: aBMD, areal bone mineral density; Adj. R2, adjusted coefficient of simple dimensional analysis, it can be argued that unlike
determination; BMC, bone mineral content; BMD, bone mineral density; DXA, maximal force, maximal stress is approximately independent of
dual-energy X-ray absorptiometry; FE, finite element; HR-pQCT, high-resolution
peripheral computer tomography. For DXA, the predictions of both maximum body size and weight,69 and therefore reflects a more objective
force and apparent stress are reported. factor of risk.

Table 3 Overview of previous validation experiments using clinical CT-based FE models to predict strength of the distal radius, thoracolumbar vertebrae and proximal
femur under typical fracture-loading conditions: axial compression on the distal radius and vertebral sections and fall on the side on the proximal femora

Structureconfiguration Age N Loading Maximum FE model Image R2 Reference

condition (Years) No. of rate force Structure, element, size, resolution ()
FM (mm min  1) (kN) w/o cortex, symmetry, (mm)
linearity

Wristpotted embalmed 829 33F 21M 200 0.42.5 Wrist, microFE, linear 165  165  165 0.75 Pistoia et al.76
Wristpotted embalmed 829 33F 21M 200 0.42.5 Standard section 9.1 mm 165  165  165 0.66 Pistoia et al.26
microFE, linear
Wristpotted formalin 819 50F 50M 100 B0.45.0 Standard section 9.1 mm, 82  82  82 0.73 Muller et al.27
fixed microFE, linear
Distal radius 2/3potted 8110 12F 9M 60 1.57.4 Distal radius 2/3, 82  82  82 0.87 Varga et al.17
embalmed tetrahedral 1.2 mm, w
cortex, fabric, nonlinear
Distal radius section 8114 18F 7M 5 0.954.7 Distal section 20 mm, 82  82  82 0.94 Varga et al.28
20 mmplate embalmed tetrahedral 1.3 mm, w
cortex, fabric, nonlinear
Distal radius 2/3potted 8110 12F 9M 60 1.57.4 Standard section 9.1 mm, 82  82  82 0.92 Varga et al.24
embalmed microFE, linear
T12-L5plate frozen, 4284 16 F&M 1 B15 Body, hexahedra, w 300  300  2000 0.79 Martin et al.40
few embalmed cortex, iso, nonlinear
L1-L4potted frozen 3787 7F 6M 9 B28 Body, voxel 674  674  1500 0.86 Crawford et al.39
1  1  1.5 mm, wo
cortex aniso, linear
T11-L1potted frozen 3183 12M 0.5 B15 Body, tetrahedral 351x351  1000 0.96 Imai et al.37
o2 mm, w cortex, iso,
nonlinear
T1-L5potted fresh 5497 77 F&M 1 B18 Body, voxel 1 mm, wo 1000  1000  1000 0.80* Buckley et al.38
frozen cortex, aniso, nonlinear
T12-L5 sectionplate 4482 37 F&M 5 B29 Section, voxel 1.3 mm, 391  391  450 0.78 DallAra et al.18
fresh frozen wo cortex, aniso,
nonlinear
Proximal femurpotted 5292 10F 8M 30 0.54.5 Voxel 3 mm, wo cortex, 1080  1080  3000 0.90 Keyak et al.51
fresh frozen linear
Proximal femurpotted 5169 6F 3M 6000 1.76.2 Tetrahedral o4mm, w o450  450  400 0.86 Dragomir-Daescu
fresh frozen cortex, linear/kill et al.55
Proximal femurpotted 55100 25F 15M 396 1.55.5 Tetrahedral o3 mm, w 250  250  750 0.87 Koivumaki et al.77
fixed in alcohol-formalin cortex, nonlinear
Proximal femurpotted 4696 19F 17 M 5 1.55.3 Voxel 3 mm, wo cortex, 330  330  1000 0.85 DallAra et al.19
fresh frozen nonlinear

Abbreviations: F, female; FE, finite element; M, male; microFE, microstructural FE. *designates an adjusted R2.

BoneKEy Reports | AUGUST 2013 7

 FE analysis to predict bone strength
PK Zysset et al
The hybrid variable aBMD has a level of strength predic-
tion that seems to depend on the anatomical site, load con-
figuration70 and the particular sample collection.55 These facts
emphasize the main limitations of aBMD as a surrogate
for bone strength. Although not all validation studies listed in
Table 3 report comparisons with densitometric variables, their
strength predictions are less consistent and are systematically,
often significantly, inferior to those of the FE method. Never-
theless, it should be acknowledged that aBMD in the spine and
hip requires substantially less radiation dose than QCT, costs
less and is easier to calculate than FE models.
The presented FE models share several limitations with
densitometric variables. First, the composition and mechanical
properties of the bone extracellular matrix are assumed to be
independent of anatomical site, gender and age. Although an
indentation study indicates that matrix stiffness is independent
of gender and age in the human vertebra,71 it was also shown
that microdamage increases72 and the bone toughness
decreases with age.73 Second, while trabecular architecture is
highly heterogeneous, its relationship with volume fraction and
fabric (or its optimality) is similarly assumed to be independent
of anatomical site, gender and age. However, there is still little
evidence that this is not the case.74,75 Third, only one validation
study accounted for the high strain rates characteristic of falls
on the proximal femur.55 This issue can be addressed by using
rate-dependent constitutive laws in the FE models, but these
need to be carefully identified for the relevant impact velocities.
To conclude, for the three most common osteoporotic
fracture sites, QCT-based FE analysis is a more reliable sur-
rogate of bone strength than densitometric variables. This
implies that despite the above limitations, FE analysis repre-
sents an improved basis for fracture risk assessment when
compared with densitometric variables such as aBMD when
calibrated QCT images are available. This improvement may
gain in clinical significance with the progress of imaging
technologies and FE modeling techniques, but obviously does
not address the other determinants of fracture risk such as the
occurrence of falls.
Conflict of Interest
The authors declare no conflict of interest.
Acknowledgements
This work was supported by the Vienna University of Tech-
nology and the University of Bern. We are grateful to Gerhard
Schneider, Heinz Pettermann, Robert Exler and Jarunan
Panyasantisuk for their contributions.
References
1. Prendergast PJ. Finite element models in tissue mechanics and orthopaedic implant design.
Clin Biomech 1997;12:343366.
2. Hobatho MC, Darmana R, Pastor P, Barrau JJ, Laroze S, Morucci JP. Development of a three-
dimensional finite element model of a human tibia using experimental modal analysis.
J Biomech 1991;24:371383.
3. Crolet JM, Aoubiza B, Meunier A. Compact bone: numerical simulation of mechanical
characteristics. J Biomech 1993;26:677687.
4. Ruegsegger P, Koller B, Muller R. A microtomographic system for the nondestructive
evaluation of bone architecture. Calcif Tissue Int 1996;58:2429.
5. Hollister SJ, Kikuchi N. Homogenization theory and digital imaging: a basis for studying the
mechanics and design principles of bone tissue. Biotechnol Bioeng 1994;43:586596.
6. Zysset PK, Curnier AA. 3D damage model for trabecular bone based on fabric tensors.
J Biomech 1996;29:15491558.
8 AUGUST 2013 | www.nature.com/bonekey
7. Beaupre GS, Orr TE, Carter DR. An approach for time-dependent bone modeling and
remodeling--theoretical development. J Orthop Res 1990;8:651661.
8. Huiskes R, Ruimerman R, van Lenthe GH, Janssen JD. Effects of mechanical forces on
maintenance and adaptation of form in trabecular bone. Nature 2000;405:704706.
9. Erdemir A, Guess TM, Halloran J, Tadepalli SC, Morrison TM. Considerations for reporting
finite element analysis studies in biomechanics. J Biomech 2012;45:625633.
10. Schileo E, Taddei F, Malandrino A, Cristofolini L, Viceconti M. Subject-specific finite element
models can accurately predict strain levels in long bones. J Biomech 2007;40:29822989.
11. Carpenter RD. Finite element analysis of the hip and spine based on quantitative computed
tomography. Curr Osteop Rep 2013;11:156162.
12. Keyak JH, Skinner HB. Three-dimensional finite element modelling of bone: effects of element
size. J Biomed Eng 1992;14:483489.
13. Zysset PK. Indentation of bone tissue: a short review. Osteop Int 2009;20:10491055.
14. Pahr DH, Zysset PK. From high-resolution CT data to finite element models: development of an
integrated modular framework. Comput Methods Biomech Biomed Engin 2009;12:4557.
15. Rincon-Kohli L, Zysset PK. Multi-axial mechanical properties of human trabecular bone.
Biomech Model Mechanobiol 2009;8:195208.
16. Garcia D, Zysset PK, Charlebois M, Curnier A. A three-dimensional elastic plastic damage
constitutive law for bone tissue. Biomech Model Mechanobiol 2009;8:149165.
17. Varga P, Baumbach S, Pahr D, Zysset PK. Validation of an anatomy specific finite element
model of Colles fracture. J Biomech 2009;42:17261731.
18. DallAra E, Schmidt R, Pahr D, Varga P, Chevalier Y, Patsch J et al. A nonlinear finite element
model validation study based on a novel experimental technique for inducing anterior wedge-
shape fractures in human vertebral bodies in vitro. J Biomech 2010;43:23742380.
19. DallAra E, Luisier B, Schmidt R, Kainberger F, Zysset PK, Pahr DH. A nonlinear QCT-based
finite element model validation study for the human femur tested in two configurations in vitro.
Bone 2013;52:2738.
20. Steiger JH. Tests for comparing elements of a correlation matrix. Psychol Bull 1980;87:
247251.
21. Colles A. On the fracture of the carpal extremity of the radius. Edinb Med Surg J. 1814;10:181.
Clin Orthop Relat Res 2006;445:57.
22. Baumbach SF, Schmidt R, Varga P, Heinz T, Vecsei V, Zysset PK. Where is the distal
fracture line location of dorsally displaced distal radius fractures? J Orthop Res 2011;29:
489494.
23. Varga P, Baumbach S, Pahr D, Charlebois M, Zysset P. Validation of an anatomy-specific finite
element model of Colles fracture. J Biomech 2009;42:17261731.
24. Varga P, Pahr DH, Baumbach S, Zysset PK. HR-pQCT based FE analysis of the most distal
radius section provides an improved prediction of Colles fracture load in vitro. Bone
2010;47:982988.
25. Hoffler CE, Moore KE, Kozloff K, Zysset PK, Brown MB, Goldstein SA. Heterogeneity of bone
lamellar-level elastic moduli. Bone 2000;26:603609.
26. Pistoia W, van Rietbergen B, Lochmuller EM, Lill CA, Eckstein F, Ruegsegger P. Image-based
micro-finite-element modeling for improved distal radius strength diagnosis: moving from bench
to bedside. J Clin Densitom 2004;7:153160.
27. Mueller TL, Christen D, Sandercott S, Boyd SK, van Rietbergen B, Eckstein F et al.
Computational finite element bone mechanics accurately predicts mechanical competence in
the human radius of an elderly population. Bone 2011;48:12321238.
28. Varga P, DallAra E, Pahr DH, Pretterklieber M, Zysset PK. Validation of an HR-pQCT-based
homogenized finite element approach using mechanical testing of ultra-distal radius sections.
Biomech Model Mechanobiol 2011;10:431444.
29. Vilayphiou N, Boutroy S, Szulc P, van Rietbergen B, Munoz F, Delmas PD et al. Finite element
analysis performed on radius and tibia HR-pQCT images and fragility fractures at all sites in
men. J Bone Miner Res 2011;26:965973.
30. Stein EM, Liu XS, Nickolas TL, Cohen A, Thomas V, McMahon DJ et al. Abnormal
microarchitecture and reduced stiffness at the radius and tibia in postmenopausal women with
fractures. J Bone Miner Res 2010;25:25722581.
31. Vilayphiou N, Boutroy S, Sornay-Rendu E, Van Rietbergen B, Munoz F, Delmas PD et al. Finite
element analysis performed on radius and tibia HR-pQCT images and fragility fractures at all
sites in postmenopausal women. Bone 2010;46:10301037.
32. Boutroy S, Van Rietbergen B, Sornay-Rendu E, Munoz F, Bouxsein ML, Delmas PD. Finite
element analysis based on in vivo HR-pQCT images of the distal radius is associated with wrist
fracture in postmenopausal women. J Bone Miner Res 2008;23:392399.
33. Melton 3rd LJ , Riggs BL, van Lenthe GH, Achenbach SJ, Muller R, Bouxsein ML et al.
Contribution of in vivo structural measurements and load/strength ratios to the determination of
forearm fracture risk in postmenopausal women. J Bone Miner Res 2007;22:14421448.
34. Genant HK, Guglielmi G, Jergas M (eds) Bone Densitometry and Osteoporosis. Springer-
Verlag: Berlin, Germany, 1998.
35. Maquer G, DallAra E, Zysset PK. Removal of the cortical endplates has little effect on ultimate
load and damage distribution in QCT-based voxel models of human lumbar vertebrae under
axial compression. J Biomech 2012;45:17331738.
36. Chevalier Y, Charlebois M, Pahra D, Varga P, Heini P, Schneider E et al. A patient-specific finite
element methodology to predict damage accumulation in vertebral bodies under axial
compression, sagittal flexion and combined loads. Comput Methods Biomech Biomed Engin
2008;11:477487.
37. Imai K, Ohnishi I, Bessho M, Nakamura K. Nonlinear finite element model predicts vertebral
bone strength and fracture site. Spine 2006;31:17891794.
38. Buckley JM, Loo K, Motherway J. Comparison of quantitative computed tomography-based
measures in predicting vertebral compressive strength. Bone 2007;40:767774.

39. Crawford RP, Cann CE, Keaveny TM. Finite element models predict in vitro vertebral body
compressive strength better than quantitative computed tomography. Bone 2003;33:744750.
40. Martin H, Werner J, Andersen R, Schober H-C, Schmitz K-P. Noninvasive assessment of
stiffness and failure load of human vertebrae from CT-data. Biomed Tech 1998;43:8288.
41. Chevalier Y, Pahr D, Zysset PK. The role of cortical shell and trabecular fabric in finite element
analysis of the human vertebral body. J Biomech Eng 2009;131:111003.
42. Imai K, Ohnishi I, Matsumoto T, Yamamoto S, Nakamura K. Assessment of vertebral fracture
risk and therapeutic effects of alendronate in postmenopausal women using a quantitative
computed tomography-based nonlinear finite element method. Osteop Int 2009;20:801810.
43. Graeff C, Chevalier Y, Charlebois M, Varga P, Pahr D, Nickelsen TN et al. Improvements in
vertebral body strength under teriparatide treatment assessed in vivo by finite element analysis:
results from the EUROFORS study. J Bone Miner Res 2009;24:16721680.
44. Keaveny TM, Donley DW, Hoffmann PF, Mitlak BH, Glass EV, San Martin JA. Effects of
teriparatide and alendronate on vertebral strength as assessed by finite element modeling of
QCT scans in women with osteoporosis. J Bone Miner Res 2007;22:149157.
45. Tawara D, Sakamoto J, Murakami H, Kawahara N, Oda J, Tomita K. Mechanical evaluation by
patient-specific finite element analyses demonstrates therapeutic effects for osteoporotic
vertebrae. J Mech Behav Biomed Mater 2010;3:3140.
46. Yosibash Z, Trabelsi N, Milgrom C. Reliable simulations of the human proximal femur by
high-order finite element analysis validated by experimental observations. J Biomech
2007;40:36883699.
47. Keyak JH, Fourkas MG, Meagher JM, Skinner HB. Validation of an automated method of three-
dimensional finite element modelling of bone. J Biomed Eng 1993;15:505509.
48. Bessho M, Ohnishi I, Matsuyama J, Matsumoto T, Imai K, Nakamura K. Prediction of strength
and strain of the proximal femur by a CT-based finite element method. J Biomech
2007;40:17451753.
49. Keyak JH. Improved prediction of proximal femoral fracture load using nonlinear finite element
models. Med Eng Phys 2001;23:165173.
50. Cody DD, Gross GJ, Hou FJ, Spencer HJ, Goldstein SA, Fyhrie DP. Femoral strength is better
predicted by finite element models than QCT and DXA. J Biomech 1999;32:10131020.
51. Keyak JH, Rossi SA, Jones KA, Skinner HB. Prediction of femoral fracture load using
automated finite element modeling. J Biomech 1998;31:125133.
52. Ota T, Yamamoto I, Morita R. Fracture simulation of the femoral bone using the finite-element
method: how a fracture initiates and proceeds. J Bone Miner Metab 1999;17:108112.
53. Keyak JH, Rossi SA, Jones KA, Les CM, Skinner HB. Prediction of fracture location in the
proximal femur using finite element models. Med Eng Phys 2001;23:657664.
54. Johnell O, Kanis JA, Oden A, Johansson H, De Laet C, Delmas P et al. Predictive value of BMD
for hip and other fractures. J Bone Miner Res 2005;20:11851194.
55. Dragomir-Daescu D, Op Den Buijs J, McEligot S, Dai Y, Entwistle RC, Salas C et al. Robust
QCT/FEA models of proximal femur stiffness and fracture load during a sideways fall on the hip.
Ann Biomed Eng 2011;39:742755.
56. Keaveny TM, Hoffmann PF, Singh M, Palermo L, Bilezikian JP, Greenspan SL et al. Femoral
bone strength and its relation to cortical and trabecular changes after treatment with PTH,
alendronate, and their combination as assessed by finite element analysis of quantitative CT
scans. J Bone Miner Res 2008;23:19741982.
57. Orwoll ES, Marshall LM, Nielson CM, Cummings SR, Lapidus J, Cauley JA et al. Finite element
analysis of the proximal femur and hip fracture risk in older men. J Bone Miner Res
2009;24:475483.
58. Keyak JH, Koyama AK, LeBlanc A, Lu Y, Lang TF. Reduction in proximal femoral strength due
to long-duration spaceflight. Bone 2009;44:449453.
BoneKEy Reports | AUGUST 2013
FE analysis to predict bone strength
PK Zysset et al
59. Lewiecki EM, Keaveny TM, Kopperdahl DL, Genant HK, Engelke K, Fuerst T et al. Once-
monthly oral ibandronate improves biomechanical determinants of bone strength in women with
postmenopausal osteoporosis. J Clin Endocrinol Metab 2009;94:171180.
60. Amin S, Kopperdhal DL, Melton 3rd LJ, Achenbach SJ, Therneau TM, Riggs BL et al.
Association of hip strength estimates by finite-element analysis with fractures in women and
men. J Bone Miner Res 2011;26:15931600.
61. Keyak JH, Sigurdsson S, Karlsdottir G, Oskarsdottir D, Sigmarsdottir A, Zhao S et al.
Male-female differences in the association between incident hip fracture and proximal femoral
strength: a finite element analysis study. Bone 2011;48:12391245.
62. Keaveny TM, McClung MR, Wan X, Kopperdahl DL, Mitlak BH, Krohn K. Femoral strength in
osteoporotic women treated with teriparatide or alendronate. Bone 2012;50:165170.
63. Keaveny TM, Kopperdahl DL, Melton 3rd LJ, Hoffmann PF, Amin S, Riggs BL et al.
Age-dependence of femoral strength in white women and men. J Bone Miner Res
2010;25:9941001.
64. Damsgaard M, Rasmussen J, Christensen ST, Surma E, de Zee M. Analysis of musculoskeletal
systems in the AnyBody Modeling System. Simul Model Pract Th 2006;14:11001111.
65. Eswaran SK, Gupta A, Adams MF, Keaveny TM. Cortical and trabecular load sharing in the
human vertebral body. J Bone Miner Res 2006;21:307314.
66. Fields AJ, Eswaran SK, Jekir MG, Keaveny TM. Role of trabecular microarchitecture in whole-
vertebral body biomechanical behavior. J Bone Miner Res 2009;24:15231530.
67. Wolfram U, Wilke HJ, Zysset PK. Valid micro finite element models of vertebral trabecular bone
can be obtained using tissue properties measured with nanoindentation under wet conditions.
J Biomech 2010;43:17311737.
68. Hayes WC, Piazza SJ, Zysset PK. Biomechanics of fracture risk prediction of the hip and spine
by quantitative computed tomography. Radiol Clin N Am 1991;29:118.
69. Biewener AA. Musculoskeletal design in relation to body size. J Biomech 1991;24(Suppl 1):
1929.
70. Dallara E, Luisier B, Schmidt R, Pretterklieber M, Kainberger F, Zysset P et al. DXA predictions
of human femoral mechanical properties depend on the load configuration. Med Eng Phys
2013;S1350-4533:001010010.
71. Wolfram U, Wilke HJ, Zysset PK. Rehydration of vertebral trabecular bone: influences on its
anisotropy, its stiffness and the indentation work with a view to age, gender and vertebral level.
Bone 2010;46:348354.
72. Schaffler MB, Choi K, Milgrom C. Aging and matrix microdamage accumulation in human
compact bone. Bone 1995;17:521525.
73. Nyman JS, Makowski AJ. The contribution of the extracellular matrix to the fracture resistance
of bone. Curr Osteop Rep 2012;10:169177.
74. Homminga J, McCreadie BR, Weinans H, Huiskes R. The dependence of the elastic
properties of osteoporotic cancellous bone on volume fraction and fabric. J Biomech
2003;36:14611467.
75. Gross T, Pahr DH, Zysset PK. Morphology-elasticity relationships using decreasing fabric
information of human trabecular bone from three major anatomical locations. Biomech Model
Mechanobiol 2013;12:793800.
76. Pistoia W, VanRietbergen B, Lochmuller EM, Lill CA, Eckstein F, Ruegsegger P. Estimation of
distal radius failure load with micro- finite element analysis models based on three-dimensional
peripheral quantitative computed tomography images. Bone 2002;30:842848.
77. Koivumaki JE, Thevenot J, Pulkkinen P, Kuhn V, Link TM, Eckstein F et al. CT-based finite
element models can be used to estimate experimentally measured failure loads in the proximal
femur. Bone 2012;50:824829.
9